-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_utils.py
331 lines (241 loc) · 9.34 KB
/
train_utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
from cgi import test
from cmath import isnan
import copy
import tqdm
import logging
import pickle
import torch
import torch.nn.functional as F
import torch.optim as optim
import numpy as np
import pandas as pd
from data_utils import state_dim
from model import Model
import ipdb
def _get_loss_with_nan(y, y_hat):
mask = 1 - torch.isnan(y).float()
y = torch.nan_to_num(y)
loss = F.mse_loss(y, y_hat, reduction='none')
loss = torch.sum(loss * mask) / (torch.sum(mask) + 1e-3)
return loss
def run_one_epoch(model, dataloader, optimizer=None, desc='', do_interpolation=False, use_state=False, **kwargs):
running_loss = 0.0
res = {
'df': []
}
def _save_to_dataframe(xy, predictions, label, interpolations, patient_id):
ts = xy.select(-1, 0).squeeze(dim=0)
ts = ts.data.cpu().numpy()
predictions = predictions
label = label.squeeze(dim=0)
label = label.data.cpu().numpy()
predictions = predictions.data.cpu().numpy()
interpolations = interpolations.data.cpu().numpy()
infer_start_tm = ts[kwargs['lookback']]
ts = ts[1:]
finger_df = pd.DataFrame(
{
'ts': ts,
'label': label,
'prediction': predictions
}
)
cgm_df = pd.DataFrame(
{
'ts': np.arange(ts[0], ts[-1] + 1),
'cgm': interpolations
}
)
df = cgm_df.merge(finger_df, on='ts', how='left')
df['infer'] = df['ts'].map(lambda x: 1 if x >= infer_start_tm else 0)
df['patient_id'] = patient_id
res['df'].append(df)
with tqdm.tqdm(dataloader) as bar:
for bid, (x, y) in enumerate(bar):
bar.set_description(desc)
x, y = x.cuda(), y.cuda()
xy = torch.cat([x, y], dim=1)
if use_state:
xy, states = xy[:, :, :-state_dim], xy[:, :, -state_dim:]
else:
states = None
# filename
patient_id = xy.select(-1, -1).mean().item()
xy = xy[:, :, :-1]
# mask all the NaN values for finger_stick
mask = 1 - torch.isnan(xy.select(-1, 1)).float()
mask[:, x.size(1):] = 0
label = xy.squeeze(dim=0)[:, 1]
# remove the first point
label = label[1:]
if use_state:
states = states.squeeze(dim=0)
states = states[1:]
if do_interpolation:
predictions, states_hat, interpolations = model(xy, mask, True)
_save_to_dataframe(xy, predictions, label,
interpolations, patient_id)
else:
predictions, states_hat = model(xy, mask)
loss = _get_loss_with_nan(label, predictions)
if use_state and (~torch.isnan(states)).sum() > 0:
state_loss = _get_loss_with_nan(states, states_hat)
# loss = loss + state_loss * 0.1
loss = loss + state_loss * 5e-3
if optimizer is not None:
optimizer.zero_grad()
loss.backward()
optimizer.step()
running_loss += loss.item()
bar.set_postfix(loss=running_loss/(bid+1))
res['loss'] = running_loss
return res
def train(args, input_dim, train_dataloader, test_dataloader, use_state=False, base_model=None, disable_train=False):
# input_dim: fea_cols + filename
if base_model is None:
model = Model(input_dim=input_dim, hid_dim=args.latent_dim)
model.cuda()
else:
model = copy.deepcopy(base_model)
model.cuda()
# model.train()
optimizer = optim.Adam(model.parameters(), lr=1e-3)
best_loss = np.inf
best_model = None
best_res = None
for eid in range(args.epochs):
print('Epoch {}'.format(eid))
if not disable_train:
model.train()
train_res = run_one_epoch(
model, train_dataloader, optimizer, desc='Train', use_state=use_state, lookback=args.lookback
)
model.eval()
test_res = run_one_epoch(
model, test_dataloader, optimizer=None, desc='Test', do_interpolation=True, use_state=use_state, lookback=args.lookback
)
if disable_train:
best_model = copy.deepcopy(model)
best_res = test_res
torch.save(best_model, f'./results/model/{args.exp}_{int(args.frac)}.model')
pickle.dump(
best_res,
open(f'./results/pickles/result_{args.exp}_{int(args.frac)}.pkl', 'wb')
)
break
if test_res['loss'] < best_loss:
best_loss = test_res['loss']
best_model = copy.deepcopy(model)
best_res = test_res
if args.exp == 'expert':
torch.save(best_model, f'./results/model/{args.exp}.model')
pickle.dump(
best_res,
open(f'./results/pickles/result_{args.exp}.pkl', 'wb')
)
else:
torch.save(best_model, f'./results/model/{args.exp}_{int(args.frac)}.model')
pickle.dump(
best_res,
open(f'./results/pickles/result_{args.exp}_{int(args.frac)}.pkl', 'wb')
)
return best_model, best_res
def run_one_epoch_union(model, dataloader, optimizer=None, desc='', do_interpolation=False, **kwargs):
model, expert_model = model
running_loss = 0.0
res = {
'df': []
}
def _save_to_dataframe(xy, predictions, label, interpolations, patient_id):
ts = xy.select(-1, 0).squeeze(dim=0)
ts = ts.data.cpu().numpy()
predictions = predictions
label = label.squeeze(dim=0)
label = label.data.cpu().numpy()
predictions = predictions.data.cpu().numpy()
interpolations = interpolations.data.cpu().numpy()
infer_start_tm = ts[kwargs['lookback']]
ts = ts[1:]
finger_df = pd.DataFrame(
{
'ts': ts,
'label': label,
'prediction': predictions
}
)
cgm_df = pd.DataFrame(
{
'ts': np.arange(ts[0], ts[-1] + 1),
'cgm': interpolations
}
)
df = cgm_df.merge(finger_df, on='ts', how='left')
df['infer'] = df['ts'].map(lambda x: 1 if x >= infer_start_tm else 0)
df['patient_id'] = patient_id
res['df'].append(df)
with tqdm.tqdm(dataloader) as bar:
for bid, (x, y) in enumerate(bar):
bar.set_description(desc)
x, y = x.cuda(), y.cuda()
xy = torch.cat([x, y], dim=1)
patient_id = xy.select(-1, -1).mean().item()
xy = xy[:, :, :-1]
# mask all the NaN values
mask = 1 - torch.isnan(xy.select(-1, 1)).float()
mask[:, x.size(1):] = 0
label = xy.squeeze(dim=0)[:, 1]
# remove the first point
label = label[1:]
if do_interpolation:
predictions, states, interpolations = model(xy, mask, True)
_save_to_dataframe(xy, predictions, label,
interpolations, patient_id)
else:
predictions, states = model(xy, mask)
predictions_e, states_e = expert_model(xy, mask)
loss = _get_loss_with_nan(label, predictions)
if optimizer is not None:
loss_aux = _get_loss_with_nan(
predictions, predictions_e.detach()
)
loss_state = _get_loss_with_nan(states, states_e.detach())
disc_loss = loss_aux + loss_state * 5e-3
# disc_loss = loss_aux * 5e-3 + loss_state * 5e-3
loss = loss + disc_loss
if optimizer is not None:
optimizer.zero_grad()
loss.backward()
optimizer.step()
running_loss += loss.item()
bar.set_postfix(loss=running_loss/(bid+1))
res['loss'] = running_loss
return res
def train_union(args, input_dim, train_dataloader, test_dataloader, expert_model):
model = Model(input_dim=input_dim, hid_dim=args.latent_dim)
model.cuda()
expert_model.cuda()
expert_model.eval()
optimizer = optim.Adam(model.parameters(), lr=1e-3)
best_loss = np.inf
best_model = None
best_res = None
for eid in range(args.epochs):
print('Epoch {}'.format(eid))
model.train()
train_res = run_one_epoch_union(
(model, expert_model), train_dataloader, optimizer, desc='Train', lookback=args.lookback
)
model.eval()
test_res = run_one_epoch_union(
(model, expert_model), test_dataloader, optimizer=None, desc='Test', do_interpolation=True, lookback=args.lookback
)
if test_res['loss'] < best_loss:
best_loss = test_res['loss']
best_model = copy.deepcopy(model)
best_res = test_res
torch.save(best_model, f'./results/model/{args.exp}_{int(args.frac)}.model')
pickle.dump(
best_res,
open(f'./results/pickles/result_{args.exp}_{int(args.frac)}.pkl', 'wb')
)
return best_model, best_res